Version 2 of the LoRA, v1 was only trained on 1 repeat and with less images.





Support Bonifasko here:
Always credit Bonifasko when posting images generated with this LoRA.
Trained on 100% publicly available images taken from e621.
Please do not sell this LoRA, or images generated with this LoRA (additionally, do not sell access to this LoRA, or access to images generated with this LoRA, e.g. via Patreon or other paid/subscription services).
Please do not remove metadata from images generated with this LoRA.
BaseModel: CoffeeNSFW
Epoch: 8
Repeats: 5
Resolution: 512
Image: 274 from e621
Runs: 9880
ClipSkip: 2
Trigger Words: None needed
Weight: Around 0.9 yielded good results
Sampler: DPM++ 2M Karras
Base Resolutions: Around 384x512
Upscalers:
4x_fatal_Anime_500000_G showed the best results
4x_fooldhardy_Remacri also did OK, but more washed out
Extensions:
Use Dynamic Thresholding Extension to use high CFGs without burn-in
Version 2 of the LoRA, v1 was only trained on 1 repeat and with less images.
1. 转载模型仅供学习与交流分享,其版权及最终解释权归原作者。
2. 模型原作者如需认领模型,请通过官方渠道联系海艺AI工作人员进行认证。我们致力于保护每一位创作者的权益。 点击去认领
